Neural Correlates of Fast Pupil Dilation in Nonhuman Primates: Relation to Behavioral Performance and Cognitive Workload

Animals

Four adult NHPs (male rhesus monkeys - Macaca mulatta), weighing (8.0-11.0 kg) were employed as subjects, they were housed in stainless steel cages in temperature and humidity controlled colony rooms with lighting maintained on a 0600:1800 day:night cycle and fed a daily diet of monkey chow supplemented by fresh fruit. All procedures were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Wake Forest University, and performed in accordance with established practices as described in the National Institutes of Health and Public Health Service policy on humane care and use of laboratory animals.

Behavioral Training and Testing

Each monkey was trained to perform a visual delayed-match-to-sample (DMS) task for juice reinforcement and tested daily 5 times per week. Animals were seated in a primate chair 1.5 meters in front of an LCD-front-projection screen and performed 150–300 trials per 60-90 min DMS test session [10,16,35]. They were trained to move a cursor into 25 cm clipart images on a 1.0 meter × 1.0 meter video projection screen by positioning the hand on a counter mounted to the chair, within a two dimensional coordinate system that was video tracked by a fluorescent marker attached to the back of hand. The DMS task consisted of a Sample and a Match phase in which an image presented in Sample phase was responded to and then a delay period of 1-90 s duration (selected at random) interspersed with the screen blanked and only the cursor illuminated. At termination of the delay interval, 1-5 non-match or “distracter” images were displayed together with the original (Sample) image constituting the Match phase of the task in which placement of the cursor into the correct “Match” phase image (i.e. same image as presented in the Sample phase) produced an immediate juice reward via a sipper tube positioned next to the animals’ mouth. All displayed clip art images were unique to a particular trial, and no image was used on more than one trial within an entire session. Animals received a portion of their normal daily fluid intake (60-100 ml/kg) during the behavioral testing session followed by the remainder upon return to the home cage. All animals were trained to a stable baseline performance level of 70-80% correct over all trials, however, as shown previously performance accuracy varied directly with difficulty or “cognitive workload” determined directly by the number of images presented in the Match phase [10]. Sets of clip art images were routinely changed frequently to maintain the trial-unique feature of each session.

Pupillometry

Due to the varying visual stimuli of the task, a pupillometry analysis that was not affected by background luminosity was required for this analysis. The analysis used here was first reported and described by Marshall at al. [27,29,46,50], applied to human studies. The technique identifies pupil dilations with rapid rise times (approx 100 ms) associated with recognizing or categorizing visual stimuli, and has previously been shown to correlate to conditions of increased cognitive effort in humans [26,28,29].

Eye movements and pupil diameter were continuously monitored from NHPs performing the DMS task (Figure 1) using video recording (EyeLink 1000, SR Research, Ontario, Canada) from an IR reflective glass (Figure 1A) placed at a 45 ° angle between the monkey and the computer display. Each animal’s head was fixed in position during DMS task performance to allow consistent recording of the right eye. Software provided with the ‘EyeLink’ Oculometry system digitized images of the pupil at 250-1000 samples/sec (Figure 1A), computed pupil diameter and direction of gaze, and categorized saccades, blink and gaze fixation events (Figure 1B). An analysis “waveform” was constructed from continuous raw pupil diameter recordings, with saccades and blinks removed (shaded regions, Figure 1B), leaving only the recordings during fixation of eye position. The resulting pupil diameter record was then decomposed using a Daubechies DB8 wavelet (Figure 1C). The “Level 1 Details” of decomposition represent fast (< 100 ms) changes in pupil diameter corresponding to the onset of pupil dilation. The Level 1 Details were thresholded to retain only pupil dilation events resulting in greater than 10% change in pupil diameter (Figure 1D). The frequency of pupil dilation events was computed as number of fast dilations/sec and compared to other DMS task events. Individual timestamps of fast dilation events derived from the wavelet analysis were recorded and used for correlation with task-related events. Since dilation artifacts could be induced by blinks and saccades, the pupil diameter was filtered prior to wavelet analysis to exclude the influence of saccades or blinks.

Surgery

Four male rhesus monkeys were surgically prepared via craniotomy and for access cylinders (25 mm anterior to interaural line, 10 mm lateral) . Craniotomy sites for neural recording were selected to overlie the stereotaxic coordinates (25 ± 3 mm anterior to interaural line, 11 ± 3 mm lateral midline) of the superior arcuate sulcus of the frontal cortex (FCx, areas 6 – and 8 – the frontal eye fields) [6] as certified from a monkey brain atlas [34] and normalized by individual MRIs for each monkey. All surgical procedures related to electrode positioning, recording and maintenance were performed under sterile conditions. Animals were sedated with ketamine (10 mg/kg), intubated, and maintained throughout surgery with isoflurane (1-3%, to effect) and oxygen. An access cylinder (Crist Instruments) was placed on the skull above the craniotomy site and fixed to the skull with screws and dental cement to allow daily placement of microelectrodes into the FCx. Each animal received antibiotics (Baytril, 25 mg/kg) for 7 days after surgery, and buprenorphine (0.7 mg/kg) as needed for pain. Animals were allowed to recover from the procedure for 10–15 days before resuming testing.

Electrode Positioning and Recording

At the beginning of daily experiments, the recording cylinder was opened, cleaned and disinfected. Recording electrodes (2-4 etched 200 μm tungsten wires, 5-10 μm tip diameter) were positioned in the center of the craniotomy cylinder such that electrode tips would enter the brain within ± 3 mm of stereotaxic coordinates for Area 8 of FCx (25 mm anterior, 11 mm lateral, [34]). The electrodes were introduced into brain and slowly advanced (0.25 mm/minute) via an Alpha-Omega Electrode Positioning System or Narishige micromanipulator to the appropriate depth for FCx recording (1-5 mm ventral to surface of brain) while neural activity was monitored to determine final positioning. Identification of recorded brain regions was based on MRI-confirmed stereotaxic placement of the craniotomy cylinders, and electrode tracks within those cylinders, which were compared to appropriate coordinates in an NHP stereotaxic atlas [34]. It was not possible to mark every cell recording locus with a microlesion because of possible damage from the number of cells recorded from the same region in successive sessions as well as expected utilization of the same NHPs in other laboratory studies. However, histological analyses in some animals confirmed recording tracks in the superior arcuate sulcus consistent with Area 8. Once the electrode was positioned in a cortical cell layer as determined by increased spontaneous firing, single neurons were isolated electronically and recorded during a 60-90 minute behavioral testing session. Only FCx neurons that did not exhibit responses to saccades were tested in the DMS task. Upon completion of the recording session, electrodes were removed, and the recording cylinder disinfected and sealed.

Single neuron waveforms were discriminated and analyzed with a Multineuron Acquisition Processor (Plexon, Inc.) and NEX software (NeuroExplorer, Inc.). Perievent histograms (PEHs) and rastergrams of cortical neuron activity were constructed for Sample and Match phases of the task in addition to histograms synchronized to wavelet-identified fast dilation events via timestamps of each recorded neural waveform. Standard scores (z = [peak firing rate - baseline firing rate]/ [standard deviation of baseline firing rate]) were calculated for each event. Cell types were identified by the presence (or absence) of significant (z > 3.09, p<0.001) increases or decreases in peak or sustained firing the PEHs [16] related to both DMS and pupil dilation events.

Effects of Sleep Deprivation

Previous studies from this laboratory demonstrated that sleep deprivation increases cognitive workload as measured by choice accuracy and response latency measures of DMS performance [10]. Sleep deprivation consisted of 30-36 hours of continuous wakefulness, verified using EEG recordings of sleep architecture, supervised continuously by laboratory personnel. Animals were maintained awake using methods described previously [10,35]. After testing animals were returned to their home cage to resume the normal daily routine. There were no residual effects of the 30-36 hr sleep deprivation regimen on DMS performance the day afterward sleep deprivation testing at which time animals had returned to their normal sleep cycles. A minimum interval of 10-14 days, mandated by IACUC regulations, was interspersed between sleep deprivation episodes for each animal. To provide an intermediate manipulation for increasing cognitive load in the task, DMS testing was also conducted on different occasions in the late afternoon when animals were “sleepy” at a time immediately prior to onset of the usual night cycle with lights off at 1600 hrs.

Neural Correlates of Pupil Dilation

Four NHPs, previously trained in the above-described multi-object, visual delayed match to sample (DMS) task [10,16], were tested following positioning of recording electrodes in the FCx (see Surgery). Figure 1 illustrates the pupillometry recording and wavelet analysis used to identify fast pupil dilation events during behavioral performance as described above in Methods. Figure 2 shows firing of a single FCx neuron synchronized to each pupil dilation event (top traces) that occurred during 100 trials within a single DMS session (as shown in Figure 2B). The traces at the top of Figure 2A illustrate the change in pupil diameter (pixels) synchronized to onset of the pupil dilation (vertical dashed line) identified by the wavelet analysis (Figure 1C). Wavelet synchronized mean firing rate increased from a baseline of 1.6 ± 0.3 Hz at 250 ms prior to dilation onset, to a mean rate of 3.6 ± 0.5 Hz at 250 ms following dilation, with a peak firing rate of 3.9 ± 0.7 Hz. Both peak and mean firing rates were significantly different from baseline (Peak: Z=7.7, p<0.001; difference between post dilation and baseline rate: F_(1,284) = 16.21, p<0.001). Figure 2B shows that both frequency of dilations and neural firing peaked soon after image presentation in both Sample and Match phases, and steadily declined until the respective behavioral response. Frequency of Dilation and neural firing rate were significantly correlated across this interval (r=64, F_(19,284) = 3.26, p<0.001).

Dilation related activity was recorded from FCx neurons in each of the four NHPs during performance of the DMS task over 10-15 sessions (n=101 cells). Figure 3 shows responses of FCx neurons classified by significant changes in firing frequency during the occurrence of pupil dilations identified by the wavelet analysis (Figure 1D), and sorted into categories on the basis of increased or decreased firing in relation to onset of fast pupil dilation. The plot in the upper half of Figure 3A depicts individual mean firing patterns as overlapping line tracings (Individual Neurons) for a set of FCx neurons (n=28) that exhibited significant increases (F_(19,173) = 4.06, p<0.01) in mean (±SEM) firing rate 250 ms prior to the onset of pupil dilation (lower histogram). Figure 3B shows the firing patterns and mean firing rates of FCx neurons (n=24) that exhibited significant brief increases in firing (F_(19,151) = 3.55, p<0.001) 100ms after dilation onset (Dilation ONSET) with a smaller secondary peak at 200ms (F_(19,151) = 2.12, p<0.01) after dilation onset (asterisks in Fig 3B). Figure 3C shows significantly elevated mean firing rate after onset of pupil dilation (F_(19,113) = 5.39, p<0.001) in a different set of FCx neurons (n=16) that maintained significantly elevated firing levels for 500-700 ms before returning to baseline firing rate (sustained, Sust.). A final group of FCx neurons (n=17) shown in Figure 3D (Inhibited) exhibited significant reductions or inhibition of firing rate relative to baseline (F_(19,113) = 3.18, p<0.001) at the onset of pupil dilation, for approximately 100 ms then reversed and increased rate toward baseline firing levels.

FCx Neurons in Different Categories Fire during Separate Phases of Pupil Diameter Change

The mean firing patterns of FCx neurons in the 4 categories shown in Figure 3A-D (colored traces) are plotted in Figure 3E along with the corresponding mean change in pupil diameter (black) before and after (± 0.4s) their respective dilation episodes. The mean firing rate traces for the 4 different categories of neurons are shown in Figure 3E (Pre, Onset, Sust., Inhib.) paired with the mean pupil diameter changes associated with the firing of that type of neuron. The respective neural discharge patterns correspond to different phases (i.e. types of) pupil dilation in that Pre and Sust. neurons were associated with similar types of dilations from a steady baseline, while Onset neurons were associated with more rapid but briefer dilations and Inhib. neuron inhibition was associated with dilations that immediately followed a brief (0.5 s) constriction.

Correlation of Frontal Cortical Activity during DMS Task Performance with Pupil Dilation

As expected from prior investigations [10,16,35] the firing of neurons in FCx increased to presentation of the Sample and Match images during DMS performance. The rasters and histograms in Figure 4 illustrate the frequency of pupil dilation events (upper) and firing (lower) during the Sample (left) and Match phases (center and right) synchronized to image presentation (solid vertical line at 0.0 s) for an Onset FCx neuron. In the Sample phase (1-image)the frequency of fast pupil dilation events increased significantly from 2.2 ± 0.4 Hz (0.5 sec pre-Sample baseline) prior to presentation to 5.3 ± 0.7 Hz (0-0.5 sec) following Sample presentation (F_(1,97) = 8.91, p<0.01), then declined to 4.1 ± 0.4 Hz within 500 ms. Firing of the this Onset neuron significantly increased from a pre-Sample baseline of 1.2 ± 0.3 Hz to a peak firing of 6.2 ± 0.6 Hz (F_(1,123) = 12.4, p<0.001) 0.5 sec after Sample presentation and declined back to baseline firing rate within 400 ms (Figure 4A). On trials with 3 images in the Match phase the mean frequency of pupil dilations significantly increased from pre-presentation baseline (1.8 ± 0.8 Hz) to a peak level (7.9 ± 0.7 Hz) immediately after image presentation (F_(1,97) = 11.17, p<0.001), then declined significantly (4.4 ± 0.6 Hz, F_(1,97) = 9.48, p<0.001) over the next 600ms to baseline levels (Figure 4B). The firing rate of the same Onset neuron increased significantly higher (relative to the same baseline) to a peak of 9.4 ± 0.7 Hz (F_(1,123) = 15.7, p<0.001) following presentation of 6 images in the Match phase, and declined rapidly to baseline firing level (1.3 ± 0.5 Hz). In Match phases with 6 images, pupil dilation frequency increased from pre-presentation baseline (2.1 ± 0.7 Hz) to a peak level (9.1 ± 0.8 Hz) immediately after image presentation (F_(1,97) = 13.42, p<0.001), then declined slightly (5.8 ± 0.8 Hz, F_(1,97) = 7.19, p<0.01) but maintained elevated frequency relative to baseline (Figure 4C). Onset neuron firing rate also significantly increased (relative to the same baseline) to a peak of 10.7 ± 0.8 Hz (F_(1,123) = 18.4, p<0.001) following presentation of 6 images in the Match phase, and declined rapidly to baseline firing level (1.5 ± 0.5 Hz) prior to occurrence of the behavioral response in the Match phase. The histograms below in Figure 4A-C show that the increase in Onset neuron firing that occurred during the trial was also synchronized to the pupil dilation events (rasters in Figure 4 A-C) that occurred on the same trials, and was thus correlated with image presentation in the different task phases (Sample vs. Match) as well as pupil dilation events that occurred on the same trials.

Figure 5 shows similar rasters and histograms for a Sust. FCx neuron in the Sample (Figure 5A), 3-Image Match (Figure 5B) and 6-Image Match phase (Figure 5C) of DMS trials. While the maximal pupil dilation frequency increased from 4.9 ± 0.6 Hz to 9.2 ± 0.8 Hz (F_(1,97) = 14.40, p<0.001) as cognitive load increased from 1 to 6 images, the maximal firing rate of Sust. neurons decreased from 10.2 ± 0.8 Hz to 5.8 ± 0.5 Hz (F_(1,123) = 11.76, p<0.001). Figure 5A-C shows that firing of Sust. FCx neurons was decoupled from dilation events, since Sust. neuron firing synchronized to pupil dilation decreased from 5.3 ± 0.4 Hz to 3.5 ± 0.3 Hz (F_(1,123) = 11.20, p<0.001) from Sample to 6-Image Match. Pre and Onset FCx neurons (Figure 5D, left) showed increased synchronized firing rate per dilation event correlated as a function of increases in dilation frequency of (Pre: r = 0.68, F_(1,1579) = 24.81, p<0.001; Onset: r = 0.75; F_(1,1732) = 21.17, p<0.001). In contrast, Sust. and Inhib. neurons showed decreased synchronized firing rates as dilation frequency increased producing an inverse correlation (Sust.: r = 0.83, F_(1,1140) = 19.09, p<0.001; Inhib.: r = 0.61 F_(1,1092) = 20.41, p<0.001) with pupil dilation rate (Figure 5D, right).

Pupil Dilation and Cognitive Load

Figure 6A shows that performance in terms of mean % correct responses (± SEM) significantly declined (F_(4,2948) = 5.13, p<0.001) on trials with increased (distracter) images in the Match Phase (i.e. increased cognitive load [10]), and that this was associated with a significant increase in frequency (2 Image mean ± S.E.M. = 5.6±0.2 Hz; 6 Image = 8.2 ±0.4 Hz) of pupil dilation events for the same conditions (F_(6,8467) = 4.21, p<0.001). Figure 6B shows a significant influence of number of images during the Match phase in terms of increasing the mean frequency of dilation for each of the 4 animals tested (asterisks *p<0.01, **p<0.001 compared to Sample).

Pupil dilation frequency was also computed on probe trials (unfilled circles in Figure 6A) in which 6-8 images were presented randomly on the screen for 3 seconds (i.e. equivalent to a typical Match phase) within the session without either 1) a preceding Sample phase or 2) response contingent reward delivery. The frequency of pupil dilation on Probe trials was no different than that observed to single image presentation in the Sample Phase of the task (F_(1,8467) = 2.01, N.S.), indicating that the increase in dilation during the Match phase of the task was not related to the change in illumination produced by the increase in number of images. Figure 6A also shows no significant effect of frequency of saccades (circles: F_(6,8467) = 1.03, N.S.) as a function of increased numbers (2-6) of Match images during the task, and only a slight increase on Probe trials (unfilled squares: F_(1,8467) = 6.18, p<0.01). There was no significant effect on duration of eye movements (saccades) as a function of number of Match images (F_(6,8467) = 2.75, N.S.), and therefore saccade duration is not shown in Figure 6A. The result of this dual relationship between frequency of pupil dilation and neural firing with respect to cognitive load in the DMS task is shown in Figure 6C which verifies that firing of Pre and Post neurons increased significantly (Pre: F_(5,1932) = 4.19, p<0.001; Onset: F_(5,1851) = 4.42, p<0.001) as the number of Match phase images increased, while the firing rates of Sust. and Inhib. neurons were significantly decreased (Sust.: F_(5,1243) = 4.54, p<0.001; Inhib.: F_(5,1150) = 3.76, p<0.001) under the same conditions.

Pupil Dilation and Factors that Alter Cognitive Load

Figure 6 displays DMS performance and frequency of pupil dilation for three animals tested under different alertness conditions: 1) normal non-sleep deprived (control), 2) evening sessions just before light cycle change (“sleepy”) or 3) following a single night (30-36 hrs) of sleep deprivation (SD) [10,35]. A two-way ANOVA of number of images vs. alertness status (control, sleepy or SD) showed a significant effect on performance (Figure 6: DMS) as a function of 1) number of images in decreasing DMS performance (F_{(4, 3271)} = 5.17, p<0.001), 2) state of alertness (sleepy and SD conditions) in reducing DMS performance (F_{(2, 3271)} = 6.58, p<0.001), and 3) a significant interaction between number of images and state of alertness (F_{(6, 3271)} = 3.98, p<0.001). A similar two-way ANOVA of frequency of pupil dilation revealed a significant effect of number of images with increased frequency of dilation (Figure 6, Pupil) on trials with more Match images (F_(5,3271) = 4.83, p<0.001), and also as a function of alertness (sleepy or SD) conditions (F_(2,3271) = 5.22, p<0.01). There was a significant interaction between number of images and alertness (F_(7,3271) = 3.01, p<0.01) with respect to frequency of pupil dilation (Figure 7). Changes in dilation frequency did not result from changes in blink frequency during the Sample (mean Control: 0.55 ± 0.12 Hz, Sleepy: 0.51 ± 0.08 Hz, SD: 0.59 ± 0.15 Hz, F_(2,3271) = 1.19, N.S.) or Match (mean Control: 0.62 ± 0.11 Hz, Sleepy: 0.60 ± 0.15 Hz, SD: 0.66 ± 0.12 Hz, F_(2,3271) = 1.47, N.S.) phases. Consistent with results shown in Figure 6A, there was also no significant difference in saccade frequency (all F_(2,3271) < 2.31, N.S.) under the above conditions of variations in alertness.